examples/optimization/aflow/aflow_pertqa.py

import os 
from dotenv import load_dotenv
from typing import Any, Callable 

from evoagentx.benchmark import HotPotQA,PubMedQA,PertQA
from evoagentx.optimizers import AFlowOptimizer
from evoagentx.models import LiteLLMConfig, LiteLLM, OpenAILLMConfig, OpenAILLM 


load_dotenv()
api_key = "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"
# Function to encode the image
import os
os.environ["OPENAI_API_KEY"] = api_key
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")

EXPERIMENTAL_CONFIG = {
    "humaneval": {
        "question_type": "code", 
        "operators": ["Custom", "CustomCodeGenerate", "Test", "ScEnsemble"] 
    }, 
    "mbpp": {
        "question_type": "code", 
        "operators": ["Custom", "CustomCodeGenerate", "Test", "ScEnsemble"] 
    },
    "hotpotqa": {
        "question_type": "qa", 
        "operators": ["Custom", "AnswerGenerate", "QAScEnsemble"]
    },
    "gsm8k": {
        "question_type": "math", 
        "operators": ["Custom", "ScEnsemble", "Programmer"]
    },
    "math": {
        "question_type": "math", 
        "operators": ["Custom", "ScEnsemble", "Programmer"]
    }
    
}

from evoagentx.benchmark import MedPertQA
from copy import deepcopy

import nest_asyncio
nest_asyncio.apply()


def collate_func(example: dict) -> dict:
    prompt = example["question_new"]
    problem = f"Question: {prompt}\n\nAnswer:"
    return {"problem": problem}

    

    

def main():

    llm_config = OpenAILLMConfig(model="gpt-4o-mini-2024-07-18", openai_key=OPENAI_API_KEY, top_p=0.85, temperature=0.2, frequency_penalty=0.0, presence_penalty=0.0)
    executor_llm = OpenAILLM(config=llm_config)
    optimizer_llm = OpenAILLM(config=llm_config)

    # load benchmark
    hotpotqa = PertQA(pertdata='adamson')
    import numpy as np
    np.random.seed(2024)
    out = np.random.choice(hotpotqa._train_data, size=150, replace=False)
    hotpotqa._train_data = out
    hotpotqa._dev_data = out

    # create optimizer
    optimizer = AFlowOptimizer(
        graph_path = "examples/aflow/pertqa",
        optimized_path = "examples/aflow/pertqa/optimized_adamson_update",
        optimizer_llm=optimizer_llm,
        executor_llm=executor_llm,
        validation_rounds=3,
        eval_rounds=1,
        max_rounds=20,
        **EXPERIMENTAL_CONFIG["hotpotqa"]
    )

#     # run optimization
    optimizer.optimize(hotpotqa)

    # run test 
    optimizer.test(hotpotqa) # use `test_rounds: List[int]` to specify the rounds to test 


if __name__ == "__main__":
    outlist = []
    main() 
    import pandas as pd
    dfnew = pd.DataFrame(outlist)
    dfnew.to_csv("./pertqaqa_save_adamson_udpate.csv")